Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Back to Podcast Digest
6 Hermes Agent use cases I promise will change your life
TL;DR
Alex’s top Hermes trick is
/goalplus meta-prompting — instead of vaguely saying “build me an app,” he has AI write a detailed/goalprompt first, which he says can keep Hermes working for 24+ hours on complex jobs like a 3D third-person shooter in Godot.The new Hermes kanban board turns a to-do list into autonomous execution — Alex starts each day by dumping tasks like “script a Hermes YouTube video,” “write three tweets,” and “make a report on trending AI tools” into triage, then lets Hermes assign agents and sub-agents while he handles human-only work.
Hermes can do competitor teardown research like a junior PM plus engineer — his example has Hermes browsing creatorbuddy.io, inspecting the console, identifying the stack, features, analytics events, and payment setup, then packaging it into a markdown-style report he can feed into coding agents.
A ‘memory wiki’ makes Hermes more useful over time — Alex has it build a site containing daily logs, topics discussed, and past work so both he and the agent can revisit prior conversations, effectively turning it into an automated diary and memory reinforcement system.
Tailscale turns Hermes into the admin of your whole device fleet — by connecting a phone, laptop, desktop, and iPad on a private network, he can ask Hermes to fetch files from another machine, install local LLMs remotely, or test localhost apps across devices from anywhere.
A 9:00 a.m. priority check-in is his self-improving agent loop — Hermes messages him every morning asking for his number one priority, then creates support tasks and updates its memory, which Alex says steadily makes the system more personalized and proactive.
The Breakdown
Why Hermes matters if you actually know what to do with it
Alex opens with a strong claim: Hermes Agent is the most powerful AI software right now, but “99% of people have no idea what to do with it.” He frames the whole video around six concrete workflows he personally uses, promising the end result is basically a 24/7 AI employee doing real work around the clock.
/goal becomes a monster when you stop prompting like a normal person
His first and “most underrated” feature is /goal, but he’s adamant that it only shines with meta-prompting. Instead of typing “build me an app,” he asks AI to generate the perfect /goal prompt for a detailed project — in his demo, a 3D third-person shooter in Godot — then lets Hermes ask clarifying questions and run for hours. He says this is the move for complex apps, major revisions, or long documents, and he often hands the output off to Claude Code or Codex afterward.
The kanban board is his daily handoff ritual
Next he shows Hermes’ new built-in kanban board, which you open via Hermes dashboard in the terminal. His routine is simple: write the day’s to-do list on paper, dump every AI-suitable task into triage, and let Hermes auto-assign agents and sub-agents while he handles things like paying his credit card bill himself. The pitch is very “manager mode”: give the work to Hermes in the morning, go do human work, come back to progress.
A quick HubSpot intermission, then straight into competitive intelligence
There’s a short sponsored plug for HubSpot’s free AI agents cheat sheet covering seven agent tools, setup, costs, use cases, and starter prompts. Then Alex pivots into a sharper use case: having Hermes perform full technical breakdowns of competing apps by opening a browser, clicking through flows, checking the console, and investigating the stack.
Reverse-engineering competitor apps without babysitting the browser
His example is Creator Buddy: Hermes visits creatorbuddy.io, analyzes the site, asks permission to inspect tools and libraries, and outputs a report with the stack, features, analytics events, payments, and costs. Alex’s excitement here is that this isn’t just market research — it becomes build input. Put the report into markdown, feed it to another coding agent, and you can emulate the best parts of a competitor very quickly.
The memory wiki is part diary, part long-term agent brain
Then he shows a personal “memory wiki,” a site Hermes builds that logs subjects discussed, daily work logs, and past conversations. Alex describes it as an automated diary for himself and a memory reinforcement layer for Hermes, because the agent can revisit those logs too. His setup prompt is straightforward: ask Hermes to build a site where you can click through topics and daily logs to revisit what you’ve worked on together.
Tailscale turns Hermes into the CEO of all your devices
One of his favorite use cases is using Tailscale to connect every device — phone, laptop, desktop, iPad — into a private network Hermes can operate across. That lets him retrieve forgotten files from another machine, install a local LLM on one computer while exposing a web app on another, or test localhost projects across devices. His phrase for it is memorable: Hermes becomes the “general computer administrator,” basically the CEO of your hardware.
The 9 a.m. prompt and the rough-but-real game demo
His final workflow is a morning priority prompt: every day at 9:00 a.m., Hermes asks for his number one priority, creates support tasks, and updates its memory about him. Alex says this is how the agent self-improves and gets more customized over time. He closes by checking the long-running /goal game build: it’s dark and stealthy “like Splinter Cell,” enemies can be shot and looted, and while it’s definitely not polished — one enemy even sees him through a wall — he’s clearly impressed that the bones of a playable game are there after one autonomous run.
Share
Keep Reading
Two Ways to Follow
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.
